MEP EP#63: HEBI Robotics with Dave Rollinson

Podcast Transcript

Host 1     00:10

Okay, welcome to the macro fab engineering podcast. I am your guest, Dave Rawlinson

Host 1     00:15

And we're your hosts, Parker Dohmen.

Host 3     00:17

And Steven Craig. This is episode number 63.

Host 2     00:22

Yeah, sorry for last week I said the wrong episode number.

Host 3     00:26

Oh, that's right. You said 61 instead of 62. Correct? Yeah. Yeah, we need to get better at making sure that our sheets or our revision control is up to date. So if you enjoy listening to the macro fed engineering podcast, please let others know about us. Tell a co worker, a loved one a friend or share it on social media

Host 2     00:46

That would be at macro fab on Twitter. That's right. I think it was a Facebook as well. Yeah. For those that want to suffer through cat photos and pictures of your family.

Host 3     00:56

Your grandma posting her her last meal? Oh, God. Oh, no. The meal she ate last night. Oh my gosh, yeah, that's terrible. I apologize for that. So we might reward your love by sending you a free koozie a little shameless self promotion never hurts, right? So if we will have a code word somewhere in the podcast, keep your eyes or your ears open yours,

Host 2     01:24

You will not see it.

Host 3     01:27

So we will, we will call out the code word. And if you email us at podcast at macro fab.com And tell us the code word and your address. We will ship a koozie off to you.

Host 2     01:37

Yes, the address is important because you will can we cannot email goods yet. That's right. At least not yet. Okay, so our guest this week is Dave Rawlinson, he had an awesome intro that he just pulled off. That's right. He is from heavy robotics. That's correct. That's heavy robotics. He is a robotics engineer with a PhD in robotics and a Bachelor of Science in Mechanical Engineering from Carnegie Mellon University. His thesis research advanced the control and design of articulated module snake like robots were the focus towards real world applications like urban search and rescue and industrial inspection. Whoo. pulled right off their website, like we always do.

Host 1     02:25

Oh, man, that's that's on our website. That means my bio is out of date. Wow.

Host 3     02:30

Well, then you can enlighten us and tell us what needs to be updated?

Host 1     02:35

Yeah. I think that's like that's a that's actually a pretty accurate I guess, if I were to think about a bio, that's probably like, my most tangible stuff to date.

Host 2     02:45

Compresses your entire life down into two sentences.

Host 1     02:48

Yeah. Yeah. So like, yeah, when you when you kind of take a step back and kind of compress what you work on, it can be kind of depressing. There was a thing, they did it, Carnegie Mellon, kind of right, as I was finishing up my PhD, called a three minute thesis. And it was like compress five years of work down to three minutes. And the fact that you could actually do it was in some ways kind of depressing.

Host 3     03:07

Yeah, that sounds brutal, though.

Host 1     03:09

It was yeah, it was great. No, it's um, being able to communicate is actually a it's kind of one of the side effects of being in a startup that's actually really kind of interesting as you're, you're going out to a much broader audience and trying to be able to concisely say what you're doing in a way that is both interesting and meaningful to other people is actually like a really important skill.

Host 3     03:31

Well, I might be jumping the gun a little bit here, but I've certainly gone to heavy robotics, the your website, and I think you guys have done a fantastic job of communicating what you do.

Host 1     03:42

Cool. Alright, thanks. That's, that's also tough, because we're kind of making it up as we go to so

Host 3     03:49

Well, I mean, there's there's plenty of let's just put it this way. There's, there's a really good mixture of text and video that kind of just drives home exactly what you do.

Host 1     03:58

Yeah, yeah, um, I guess I should probably just real quick say for people listening like heavy robotics. what the company does in a nutshell, is we are making smart, robotic modular building blocks. So smart, robotic actuators, what we're focusing on is basically single joints that you can put together to make any robot you want with the inspiration being that we would really like building a robot to be as easy as putting together Legos. That's the dream. That's where we're going. And where we're starting is basically a smart actuator that you give it power, you give it Ethernet, and then you're off to the races.

Host 2     04:34

So how, I guess so basically, the the hardware, the guts are done, and use apply a really good I guess, API on top of that, do you control it?

Host 1     04:43

Yeah, I would. I would say that the hardware and the guts are done for the product that you see on our website. Now, what we call the X series actuator, what we have is basically the ability to kind of wrap up motion control in a in a compact form factor, but the details that that form factor can change for every application. So it's all about kind of trying to find the right sweet spot in terms of size, power, price point, capability, a whole bunch of other things. But you're exactly right that there's there are basically two main parts and one is picking the form factor for the physical product, the actuator. And then yeah, the other part, which is really just as important is the the API's in terms of how you control these from a computer, how you coordinate them. Because there's, there's, I mean, you guys know, there's, there's, there's a ton that changes from going from an individual joint to coordinating as, as an entire robot, there's a whole kind of layer there, that is actually very difficult to get right. And that's what we're trying to solve while at the same time making flexible for people so that they can customize however they want.

Host 3     05:46

Oh, yeah, yeah, yeah, if any, if anyone hasn't done the trigonometry, but behind, like a, like a three jointed robotic arm, give it a shot it, it's nowhere near as easy as it seems.

Host 1     06:01

Yeah, that's exactly right. So they're, um, it's a whole branch of kind of basically math called kinematics. It's just basically the math of like, motion, the math of representing things in space. And as soon as you go into 3d, it gets really hairy, it gets really nasty. And that's a lot of what people wrap up for you. If you buy a robotic arm, you're kind of basically paying really smart people to have sat down and done all this math for this specific arm. And what we're trying to do is actually kind of wrap that up in a way so that you could build any arm and still get, you know, as much as we can give you most of those benefits

Host 2     06:30

For a little more complicated than an Excel spreadsheet, right?

Host 1     06:34

Yeah, probably.

Host 3     06:36

But you're actually selling these kind of modular units that act as joints, right?

Host 1     06:42

That's correct. So So yeah, what we sell is basically a joint that has all the control and everything wrapped up. It's designed so that you can pretty much just bolt it to whatever, whatever you want. We have kind of bearings built into it. So you don't have to spend too much time kind of carefully thinking about how you how you loaded how you have to support it, you know, kind of externally, the idea is that you should be able to clamp to it and can lever it. So you don't have to be too much of a monkey to really, you don't worry about breaking it. Basically. It's really robust. You can use it as a wheel, you can use it as a base. Yeah.

Host 2     07:17

Cool. Yeah. So I guess we'll wind the clock back a bit. And, like, how did you get into robotics? Because that seems to be like all you did in terms of your bio.

Host 1     07:28

Yes, that's, that's all it's on my bias. So

Host 3     07:32

I'm sure there's a lot more behind there's,

Host 1     07:34

There's there's more behind that. So I grew up in a small town in Virginia little town called Clifton Forge. And in high school, didn't really do a whole lot in robotics, but decided, for whatever reason, I wanted to do engineering, and came to Carnegie Mellon for school. And my second summer job was basically an internship here at a company in Pittsburgh called redzone robotics, and we made sewer robots like, like literally robots that went down into the sewers, into poop. And

Host 1     08:05

It's actually was about to ask.

Host 1     08:09

Um, and, and, you know, it sounds silly, but that's, that's what got me hooked on it, because I was seeing kind of this amazing technology that was being put to work in an area that needed it, that still needs it, you know, really, really desperately. So infrastructure general. And in particular pipes, where they're underground, they're aging, they're super expensive to replace, they're super expensive to maintain. And the problems only getting worse as they get older. And, and so the the breadth of things that you needed to be able to know and manage, just kind of always made things interesting on day to day, and then just kind of the motivation of really like seeing how you're pushing this technology forward to really impact the world in a real way. Got me super excited. So I'd say my work into robotics really started with that, which I guess was my sophomore ish, year of college back in like 2004. And then it's kind of grown from there. So I like I worked at Red Zone, and then basically went back to grad school in 2009, to focus more on the control side. So before, I had really been mostly mechanical engineering, focusing on really like the nuts and bolts side, so designing really robust equipment, housings, mechanisms, tools, to really stand up to both the environment of the sewers, which is just kind of a harsh, brutal environment. But also kind of like just as important to learn is the, the way people use your equipment to so you're designing these robots, they go out to the field, and the guys that actually go deploy these things, you know, they're really hard on the equipment, they're not hard on the equipment because they're, you know, being mean, it's just because that's, that's how they have to get their job done. And so kind of creating an appreciation for the fact that people, the way that people use your stuff isn't the way necessarily that you would have originally intended and having kind of like a respect for that and really kind of like a love for embracing that was also really cool, which is kind of a spirit we've tried to, I've tried to bring with me back into the heavy too, because I'm super excited to see the way people use, like our modules in our building blocks, even if what they're doing is like breaking them in new, amazing ways.

Host 2     10:12

I was about to ask them, you know, you were talking about the enclosures and stuff that you were designing for, for the sewer robots? Is there a rating for? Like, is it IP 67 rated, where you don't have to make sure it's submersible, or is there another rating for PU. A,

Host 1     10:32

Not for pu specifically, but like from a robustness standpoint, um, you know, it's IP 68 Plus, like, in the grand scheme of things, it's, it's to me as a as a Mechi. If you really, really needed to be robust in the field, like, it's got to be like, Oh, ring seal, there's kind of like, have a quick line that you cross between, like kind of sitting outside in the rain and then like, beyond that, it's pretty much gonna be good to sitting at the bottom of a swimming pool indefinitely. Like that's, that's kind of that's kind of the level, you immediately need to take it to, for to really work in the field all the time being thrashed, and bashed, and all that other stuff. So like, everything I designed was IP 68. So like ordering seals, in some cases, we were in like explosive environments. So we had to be able to do like a positive nitrogen purge and stuff like that.

Host 2     11:20

Yeah. So you're like class one div one.

Host 1     11:23

Not like I was actually not officially certified. But we were basically designing we were designing with those specs in mind. Okay. But But yeah, so that that level of kind of sealed robustness. Exactly.

Host 2     11:35

And another tangent on that same subject is like, did you ever get a device back from like, the field? And it's like, oh, clean, and you open it up? And it's filled with the sludge?

Host 1     11:46

Ah, yeah. I mean, so what was the what was interesting is we yeah, basically, I mean, we, the side of it, that we that I was on, we deployed all our own equipment. So like, we were the people that dumped our own stuff. No, okay. Bigger bit is that like, once things go in the sewer, there's a certain like film that never really ever comes off. So we would like we would borrow stuff from people and they bring it back and be like, You didn't clean it. They're like, Oh, we cleaned it. Trust trust us that's that's as good as it'll ever get.

Host 1     12:16

The metal has absorbed the property.

Host 3     12:18

Well, that sounds like great fundamentals for for getting into some some the robotics that you have it at heavy. So let's talk about the beginnings of heavy Can you give us some information on it?

Host 1     12:33

Yeah, so um, so heavy was founded by by five people. It's a how he chose it. He was a professor at CMU. And then I was one of four of the people in his lab, both staff and students, Matt Tesh, myself, Florian, enter Curtis Layton. We had all been kind of working together in the lab in one form or another for three or four years at that point. And kind of, we worked on snake robots was was kind of our main thing. And for a while we were like, man, there's there has to be an application for snake robots, there has to be an application for snake robots. And as we kept looking for, for ways we could start a company around that. What we want to finding is that for any sort of task, where you'd want to take a snake robot, so like a long modular robot that could kind of twist and turn its way through any sort of shape or terrain. By the time you got a robot that could actually do something from like a controls perspective, in that environment, you're basically better off taking that time and building a purpose built robot for that task. But the act of building that purpose built robot was also equally a pain in the butt in different ways. And as we kind of took a step back, what we realized is, rather than a snake, what we really had, were these modular building blocks that we're using to build our snake robots. And so we that's what we wound up kind of taking, and and pushing for it. So rather than take snakes and make them general purpose, take the technology that's underlying the snakes and try to make that a little bit more general purpose.

Host 3     13:58

Yeah. So like, just a box full of robotic elbows.

Host 1     14:03

Yeah, yeah, exactly. Right. So what with if you look at kind of the progression of stuff from the lab, and even some of the stuff that you'll see on our website? You know, the first modules we have we call the S series s for snake, I guess. And and you know, that there was a certain form factor of their sleek and they're small, they're sealed, but there were limited and a lot of ways. So what I what I've told people is basically we're taking that same technology that was built for snake robots, and making something that's good for anything except a snake robot, because that turns out to be a much larger market.

Host 2     14:35

Actually, go back to two to two robotic elbows. I think that will make that the code word.

Host 3     14:42

Oh, yeah. Code word robotic elbows, email podcast at macro fab.com for your Kuzey.

Host 3     14:55

So you're more than our robots. Okay, so you have right now on your, on your website, you have two different models, you have the X model and remind me what the other one is again.

Host 1     15:06

So the other one is the s, okay? Which I found out is horrible for audio, right? Because those two letters sound the same. So it's like x is an x ray and s as Sierra are the two series.

Host 3     15:16

Yeah, right. So so the, the S is in Sierra is the is the elbow style. Yeah. And the and the xylophone x is the is, is explained that it's a it's like a like a, like, a servo in a way.

Host 1     15:30

Yeah, it's servo it kind of, uh, you know, it's hard to describe it over the air, it kind of looks like a metal tape dispenser, it's kind of the shape of it, but the middle of it spins continuously. So, kind of the, whenever you're designing kind of a new widget, you pick your constraints to start with, and we kind of knew the motor we were using on the inside. And everything else was basically driven by the fact that we wanted to connect this as quickly and easily and cheaply as possible. And we already ran Ethernet. So basically, it's a, it's kind of a flat pancake form factor with an output hub that you can bolt to. And then on the sides, you basically plug in to rj 45, because you basically daisy chain Ethernet up and down to communicate to these things. And I'm really not exaggerating, like the form factor of this actuator is basically driven by the fact that we had to put two rj 45 connectors which are enormous on the on the sides of the modules and and then we wanted a through bore through the center that could pass an RJ 45 connector, or, you know, more than one after us FD string on through.

Host 3     16:30

So so your form factor was actually determined by the connector? Yeah, that's basically right. Yeah, that's great.

Host 1     16:37

Yeah. So it, it looks kind of like a little bit weird by itself. But it's turned out to be a kind of a really good form factor. Because that also making it flat and broad, you can bear a lot of force, you can bear a lot of torque. So even though some of our, like, smaller gear ratio modules really don't put out that much torque, you can still use them as like the wheels of a car, for example, because the the actual load rating at the output is quite high. And then the through bore is actually I think, pretty critical for creating flexible tasks where you have a risk that might be able to turn, you know, 234, or five times you can pass things like pneumatics through the center of it, we've really kind of, I think, hit upon that is like a pretty decent general purpose form factor, at least for for the time being. Well, that's

Host 3     17:21

Great. So you have configurable flavors of these.

Host 1     17:26

Yeah, so there's, what's on the website right now is the x five. And there's three different gear ratios kind of within within that, that that module, so there's a there's a spur gear train on the inside. And by changing out two of the stages, we can give you everything from nine Newton meter, continuous torque at 15 RPM to about one and a half newton meters of torque at about 90 rpm. One of the things we kind of were looking at as we kind of looked at the market was that a lot of people are looking at making manipulators and the the what you want in terms of torque and speed at different points in a manipulator, whether it's the base of the shoulder or the elbow, the wrist are very, very different. So we wanted to within one form factor support, kind of different gear ratios, trade offs between speed and torque. And then another one that's not on the website yet is what we call the X eight, which is literally the x five, but just kind of stretched and fatter. So it's basically the same gear ratios, but double the power, so more torque at those same speeds. So it's got a bigger motor in there, it's got a it's got a bigger motor, and then thicker gear faces to bear the higher torque. But what's cool is we actually reuse the exact same electronics. So it's the same circuit board and both more or less the same firmware in both, basically the same firmware with different configurations, runs all the same API's. So we're trying to basically kind of grow out the range starting at the bottom going up. While trying to kind of keep things manageable from a from a logistics perspective, eventually, I think we want to have about probably about 12 different skews covering basically all the speed and torque combos of kind of upper body, upper body torso, I doing tricks like that.

Host 2     19:02

So the the rotating mech, the bearing interface. Like cuz that's got to be one of the most complicated things in that. That device, the X block? Yeah. Because like you've got do because a lot of times when you design something like this, you have a known like it's going to be side load, or it's gonna be a thrust bearing or something like that. How did you go about in designing that if you can.

Host 1     19:35

So, in that one, actually, kind of the, the Eau de was really driven, the size of it was actually very much driven by the torque of the the highest gear ratio. So the forces that the teeth of the gears internally we're going to see at the highest Yuri show, kind of drove teeth of a certain size of a certain diameter. And then another thing I'll say is, all of these actuators are what we call series elastic. actuators, which is basically means that we've taken a robotic servo, which is normally very stiff, it's a motor and then a gear train. And then that's about it, what we do is we deliberately put a spring between the gear train the end of the gear train and the output. And that makes the actuator physically compliant. And it also lets you basically sense the deflection of the spring to control the torque. So a key part of these actuators is that they're not just position control, they're also torque control. So if you're familiar with robots, like Baxter or Sawyer from Rethink Robotics, basically the same technology is baked in to these actuators. And so the spring we have basically a Sony output that also kind of drives the outer diameter. And then from that, it's a matter of basically just trying to find commodity bearings that go in the outputs. So in that case, that's a, we started off with basically a 45 millimeter inner diameter, thin section ball bearing the 6709, ZZ for those of you who really no bearings, and then going forward, now that we have a little bit more kind of experience under our belt, we're actually going to be moving to a cross roller bearing for all these, because like you pointed out, like there's, there's a whole bunch of different ways that people can load things. And most industrial robots actually have what are called cross roller bearings, which are instead of having balls in a bearing like you would normally have, they're actually a bunch of cylinders that are slightly tapered cylinders that are alternated, and they're extremely rigid, and they can handle

Host 2     21:27

Oh, so it's like, it's a roller bearing. But you said they're tapered, but and crossed

Host 1     21:32

There. Yeah, they're, they're criss crossed. So think of it as I'm in a straight line, it would be a whole bunch of like, really short cylinders, so like maybe, like, millimeters wide and two millimeters long. And they alternate in terms of leaning to the left lane to the right, leaning to the left lane to the right. So they almost look braided. Yeah, yeah, exactly. Yeah. And then it now if you if you take that you put it in a circle, what you have is basically a ring of these things. And they can take thrust loads and radial loads really strong. But most importantly, they take what's called cross moment loading. So you can basically can't leave or something off to the side and hold a ton of force because the the contact points of these things are all in a line, as opposed to like a single point, theoretically, like you have the ball moving moving to that going forward, which adds some expense to the modules. But I was

Host 3     22:17

Just about to say, that sounds really cool. And extremely expensive. That is correct. So So I'm curious with this spring that's added into kind of the drive chain. Does that affect your positional accuracy? Does that kind of make it a little bit? spongy?

Host 1     22:34

Yeah, no, that's That's exactly right. So, um, the way to describe series lasticity to people is that the the compliance are kind of spurious of your system, you're deliberately making a trade off between being able to do force control well, and doing position control well, so the stiffer you are, the easier it is to control your position.

Host 3     22:56

And to stop on a dime,

Host 1     22:58

And to stop on a dime, if you know you need to stop. But the ability to control your force becomes much more difficult because you have to react much more quickly and your system responds much more, you know, much more stiffly. So you kind of think of it as like a extreme, it's like a slinky, right? Like, if I take a slinky, and I drop it down, I can hit the top of the table, and I can kind of move my hand up and down and very easily, like if it was sitting on top of a scale, I could, I have lots of room in which I could adjust my hand to kind of dial in exactly as much weight as hanging down at scale. But controlling the end of its position is a nightmare. So you know, depending upon whether you care more about position control, or force control, kind of dictates how much compliance you might want your system and kind of kind of walking that line is something you're we're still exploring really to.

Host 3     23:44

Yeah, and and Parker and I, when we both kind of watched some of your videos of the kind of widgets and gizmos that you created with these. And the whole time we're like, oh, yeah, yeah, we're gonna get these, we are going to get these and make something out of it. Because we have some, we have some projects right now that they, you know, go right, along with these kinds of devices that we're working on at macro fab right now. And so, you know, I was curious, what is your the accuracy and repeatability on these devices?

Host 1     24:18

Yeah, so the, the classic academic answer is, it depends. Um, but the, how it always is. Yeah, right. So, at a joint level, we have repeatability that is, you know, down in the arc minute range, right, like that's, that's, that's a lot of so at a joint level, we have extremely high resolution encoding on the output. And if it's unloaded, you can dial things in very accurately so but putting things into other into a system, which is what most people care about, if you're making an arm, for example, that has kind of human scale reach. And with these modules we have right now you have about a kilogram payload. You're looking at repeatability that's in you know, a range of about a millimeter like millimeter level. repeatability for kind of like a loaded system doing kind of like a real world task. But the the joints themselves have a, you know, a sub arc minute.

Host 3     25:12

Well, right, right. I mean, of course your your accuracy changes once you integrate them all into a giant system.

Host 1     25:19

And for us, it really does come down to control. So we have, we've basically built in as high resolution sensing on the position as as we can get. But the the limiting factor is basically in terms of how tight you can dial in the controls. Partly because since these things are compliant, like elastic, you can't really crank up the gains like you can on a stiff arm. So in addition to just kind of making the position control a bit more difficult, if you try to, like physically in the in controls make the arm as stiff as industrial arm, you actually start to oscillate because you know, the, the frequency at which the thing will start to jitter is much lower, because you're much more compliant. Yeah,

Host 2     26:03

I'm gonna guess you hit that spring in it, that that that encoder keeps trying to adjust the motor. Right?

Host 1     26:11

Yeah, yeah, basically, um, there's,

Host 3     26:14

Well, it shouldn't do that. Well, no, he makes this

Host 2     26:17

As he's explaining like, if you make it as tight as a rigid arm,

Host 1     26:21

Right, right. Yeah. So like, if, if, if there any Mackey's listening and they remember their you know, junior level controls, whatever resonant frequency is screwed of k over m, right? So like, the stiffer the higher the resonant frequency. And basically, you can theoretically control past the resonant frequency. But essentially, if you start, if anything non ideal happens, like I don't know, noise in your velocity readings like there always are in robots, then like going past that, you basically, you're very quickly getting a motor, you want to oscillate. So kind of like a rule of thumb, you want to stay at like kind of half your resonant frequency to avoid kind of the jitter. And for an industrial robot, you're super stiff that your resonant frequency is up in like the hundreds of hertz, it just never matters. But when you're more compliant, where our resident frequency is closer to like the 1010s of hertz.

Host 3     27:01

So let's get into that mud,

Host 1     27:04

If you if you care about being being stiff, right, so kind of like one of the things we're exploring is, you know, we industrial automation has kind of been built around the idea that I'm going to make something super stiff and super accurate. And then use that as like the sledgehammer that I'm going to apply to every single problem. And we're, we're looking at it more of Okay, what if I can control force? What if I could actually customize the mechanisms so that I need fewer degrees of freedom? What if I can use modules that have lighter payload, but can do it intermittently? What if I can incorporate other things like springs, dampers and smart ways? What if I can incorporate lots and lots of other sensing, right? So like cameras and connects and touch sensors, and tactile sensing? You know, what if I can bring all of this other stuff to bear in a really flexible way? How does that change the equation of what we can apply? And so what that means is that we're going about it in a way that like, if you try to automate a task in a traditional industrial way, we are going to fall short, because it's kind of the bet we're making in terms of technology is kind of in a different direction. So that's why I think most of our attraction right now is in research in academia, we feel that as, as a company, our big play will be in the industrial space, industrial space, somewhere, we're, but we're still kind of hunting around to find out, you know, what, what the real needs, what the right kind of niche to slot into might be before we really kind of commit to it really, really hard.

Host 2     28:21

Cool. So when you're, we're designing and testing these modules? Was there anything that was like, you know, unexpected that came up? In terms of the design, you had to make a change or something like that?

Host 1     28:33

Um, well, the answer is always. There's tons of stuff. Oh,

Host 2     28:39

Yeah. What was the one thing that's like, you can go back to? And yes, now,

Host 1     28:43

The biggest one that I will point to, and this is this is really kind of like under the hood, but there was there was, I will tell you, there was a whole version of kind of this product that never saw the light of day. And if you came out and kind of saw us at booths at like the DARPA Robotics Challenge, or maybe some some conferences in like 2015, you will have seen this thing. And what it was, in terms of I think being unexpected is like, it was this product, but any form factor that you could picture is the opposite. It was longer, it was skinnier, it didn't have a through bore connecting to it was way more clunky. So it was basically the same idea. But we just completely inverted between that product and this product, we completely inverted what we thought were the important things. So I'll say what was kind of unexpected, as we said, Okay, we have this technology that we have wrapped up for snake modules, right, we knew we wanted to make it more general purpose. So we said okay, let's make it. Let's make it have an aluminum extrusion. So we actually tried to build it into like a chunk of like 8020 or like T slot aluminum, so that you can kind of connect to it really easily. And then aside from that, we really focused on the inside making it super powerful and super accurate and super strong. But we didn't really take a step back and think about oh, you know, This isn't really useful in and of itself, we have to connect it like that it only really matters if we have three or four of these put together. And the way they might be put together is extremely different. So we were doing things like using an industrial connector on the back, which was sealed and really robust. And, but it also costs like, you know, $200 for a cable to connect these things, and we're like, it's gonna be really obnoxious if you're building a robot, and you're gonna have like 1500 bucks and cabling, just to put it together. And then the the form factor in terms of like how you mounted it, like, if you're gonna make an arm, you had to put these extra brackets around it, and then the cabling was all running around and you had service loops. And then there was, the power of it was actually so high, it actually kind of started to be a little bit dangerous, at least it seemed to us. So we basically put a big pause on that, and kind of went back to square one and said, Alright, let's go to a little bit lower power. And let's really think about how the heck this thing actually gets used in the system. And that drove us to things like I talked about where it's like, it seems silly to design it around a connector, but like, everybody can go to Amazon just get gobs and gobs of Ethernet cables and connect the robot with Ethernet cables of any length they want. So let's design around that let's use standard molex power connectors, let's give them a three bar. So the wiring isn't a nightmare. And they'll say we can make it a big flat pancake. So they can basically do whatever they want with it in terms of mounting, without really having to be like a real Mechi about it and have to understand like how the forces and loads are being handled, it's just kind of like handle all the cases. So I would say that was probably the most unexpected thing was like we thought about the actuator so much. It wound up kind of being a dead end. And what you see now was basically like a full restart on that that design process. Hmm.

Host 1     31:33

That's really interesting.

Host 3     31:35

Well, yeah, and and what I kind of see from that is, like, Take, for example, a macro fab, a lot of times we have to rapidly build a jig or come up with a solution or have something in two weeks. Yeah. You know, and, and if we're having to, you know, if we're looking for a solution, and we see that, you know, this connector is gonna take a while to get in. And it's super expensive, that goes out the window real fast. But something like this could be a solution really quickly, especially if it's something like, Oh, I just plug an Ethernet in and talk over this API. And there we go.

Host 1     32:09

Yeah, no, that's exactly right. So we want, we want to be able to, but most of the design decisions, most of the design decisions we make, are basically based on okay, what gives us more flexibility and especially what makes the development easier, or faster, right, we want to be able to get from zero to a useful system or a useful prototype very quickly. And part of that's the form factor. And part of it is also just kind of us focusing as much as we can really just on the actuation. And then kind of like the immediate steps out of what other glue, you might need to hold the system together, whether it be brackets, whether it be kind of maybe some general purpose input output, but we basically want want to make it so that you buy the actuators from us. And then we really want you to be able to get the rest of your robot from Amazon or mcmaster carr. Like basically that's, that's kind

Host 1     32:52

Of saying mcmaster carr.

Host 3     32:54

I've got two thumbs up right now. Yeah.

Host 2     32:57

Speaking of that, this goes right in our next question is supply chain. So how much do you like do you all build in house um, it sounds like you're more research so you might not have like a whole you know, factory floor people building stuff,

Host 1     33:13

We don't we don't have like a full machine shop but all the final assembly all the testing all the verification and calibration, that all happens in house. So we we outsource basically machine components and and boards for prototype stuff will we will actually assemble and bake the boards in house. But basically at that kind of at the component level, things come from all around the world. We have 3d printed parts, we have machine parts, we have extruded parts, we have motors, we have gears, almost everything's custom SPECT and outsourced and then the the all the final assembly comes is in house and I think that's we'll probably stick with that for now. For the foreseeable future unless unless like we hit like mad mad scale where I I am more and more convinced of the importance of kind of controlling the the build process as much as possible. And so like as as we grow I would actually love to bring like more machining and maybe even more capabilities in house but definitely don't want to give away like the the final assembly and the calibration just because you very quickly, it amplifies your inertia, right like so every time we build like a batch of 50 of these there's there's subtle tweaks that from the outside aren't apparent but on the inside are making big differences in both kind of like the quality of the motion, the quality of the control, the time it takes to assemble the reliability the lifetime and those are all things that get kind of really really hard if you outsource the whole thing soup to nuts. It's it's it's an interesting line to walk and we're I'm still kind of dialing in and changing my mind in terms of kind of what we want to control where. In terms of supply chain, I fully appreciate why it's called a chain right? Like it is a serial thing and every link is broken. If any link is broken, you're screwed. It's not called a supply parallel

Host 3     35:00

Well, I mean, if you're at the point where you're outsourcing the final final assembly, then your product is probably fairly mature.

Host 1     35:11

Yeah, um, but I, I do talk to a lot of startups that seem to try to like go pretty deep on the outsourcing like right from the start, which I think is kind of a, in my mind kind of a recipe for disaster. I think there are there are a little, I think, I think, when it comes to hardware, it's very easy to think that you can kind of turn turn the key easier than you can, especially if you're trying to do anything that's like pushing the envelope in any direction.

Host 1     35:43

And but I've tonight well, also, I will also say that when it comes to mechanical stuff, life is so much harder, like so our board guy Curtis, like, and for you guys, like dude circuit boards are like, so there's still a lot of room for wiggle room. But like, it's so much more standardized than like getting machine parts and stuff. It is is way more of a standardized process in terms of boards and stencils and population and design rules and gerber

Host 3     36:07

Files, we have our box of Legos, and we kind of put them on a board.

Host 1     36:12

Yeah, I'm super jealous of that process as

Host 3     36:15

Well. You have a bunch of fasteners, it's just yeah, okay.

Host 2     36:19

Sure. Yeah, I was gonna say is, is like, yeah, when you go to like, you know, a board house or whatever, you can download what they can do. And you punch that into your design tool, to make sure your stuff can be designed when you go to a machine shop to like, how many zeros? Do you want to tack on to your precision? It's like, I don't know, what's good. What do I need? Plus minus a mil? Or do I need, you know, doesn't matter? Yeah, the

Host 1     36:47

Degree to what you tell us things is as much of an artist as as it is a science and then even as you work with, with different shops, shops just have different specialties and it's way more of kind of a softer line than it is, like here are my capabilities. It's like, oh, you know, I have I have a, you know, a five axis CNC. But man, for whatever reason, I really know how to cut aluminum. Like if it's 6061 or 7075. Like you're good to go. If you need something that's like hardened steel. Well, you know, there's this other guy that likes to do it, you know, some people have wire EDM and really know how to use it. Some people, you know, have basically they'll say, EDM, but it's really just a sinker EDM that they used to, like, extreme taps that broke off. Um, so it's the Yeah, it's just, there's so many more variables.

Host 3     37:35

That would be that would be so annoying. If that was true in the, in our field, where we had to know which manufacturer could manufacture specific parts better than others. That would be Oh, that'd be terrible.

Host 1     37:48

That's exactly based on the part that I design. In my mind. I like I I basically have them quoted by different shops based on my experience and what I know that they're good at. Yeah, it's, I tell people, like if you're, if you're gonna be a mech, and you're gonna be in mechanical design, like, half of your half year, you should like there should be a class on like, knowing when to plead and when to yell at vendors, basically, like vendor relationships is actually like, like an integral integral part of being a mechanical engineer.

Host 3     38:17

Oh, sure. Sure. Not gonna shoot the shit with the with the guys at the shop, right?

Host 2     38:22

Yeah. Oh, yeah. You could roll that into like, what's the engineering ethics class? Engineering ethics and how to yell at vendors one to one. Yeah. So heavy, right? Where'd y'all come up with that name?

Host 1     38:40

Uh, so kind of, it's it's basically an homage to our snake robot roots heavy is Japanese for snake. And, okay, and it's so we wanted something that was short. So you can you can put it in front of in front of things and not have it make the words too cumbersome. And Japanese is also an appropriate language for for snakes and Snake robots because that's basically where that field started with a guy named Professor Hirose back in the 70s, who was building snake robots out of basically like transistor logic parts, like long before we came around, and we're doing doing crazier things.

Host 3     39:13

Yikes.

Host 1     39:15

That's cool name.

Host 1     39:16

Yeah, thanks.

Host 3     39:17

So what's what's the website in case our listeners want to check it out?

Host 1     39:21

The website is heavy robotics. COMM And also heavy.us If you want to take fewer characters.

Host 3     39:29

Okay, that's H E, B.

Host 1     39:32

H e bi. Yes.

Host 2     39:34

I cool. And I guess with that you want to sign us up Dave?

Host 1     39:40

Uh, yes, yes. So thank you guys. This was the macro fab engineering podcast. I was your guest Dave Rawlinson

Host 3     39:47

And we were your hosts Parker Dolan and Steven Craig. Later everyone take it easy.

Transcribed by https://otter.ai